Reactive Collision Avoidance of UAVs with Vision Sensing using Pin-hole Cameras

Using a pair of simple passive pin-hole cameras, an effective reactive collision avoidance algorithm is presented in this paper for unmanned aerial vehicles. First, an extended Kalman filter approach is proposed to extract the useful information from the noisy information generated. This formulation takes advantage of both ‘stereo vision’ as well as ‘optical flow’ signatures and hence is capable of estimating the range information as well, making its position estimate quite accurate. Next, an ‘aiming point’ is computed after putting an artificial safety ball around the obstacle and using the collision cone approach. After that, the velocity vector of the vehicle is steered away towards this aiming point using a recently developed ‘differential geometry guidance'. A large number of simulation studies, which also includes consistency checks for Kalman filtering, leads to the conclusion that this strategy is quite effective in avoiding popup obstacles within a very short time and hence can be very useful for reactive collision avoidance